Accelerating Cassandra Performance: An eCommerce Case Study

The digital shift during Covid-19 was unexpected, but many businesses have responded and adjusted quickly to the changing landscape. The retail industry in particular, which has been historically dependent on both online and in-store shopping, is adapting to the growing digital trend. 

Customer Experience has become crucial! 

With increasing web traffic combined with the need for personalized shopping experiences across digital platforms, e-retailers and eCommerce giants are focused on engaging customers by providing:

  • Real-time inventory by store
  • Curbside pickups and flexible shipping options
  • Customer reviews and product specs
  • Suggestions of similar or popular items with inventory, i.e. customized recommendations

All play a vital role in both as a revenue generator and loyalty builder for the user and a means for the customer to fully enjoy their shopping experience.   As industry leaders like Amazon have spent huge amounts of financial and engineering resources on perfecting the online shopping experience, customers now expect similar experiences even from small or midsize retailers.  In order to deliver both a top notch experience and maintain cost and complexity, retail and eCommerce organizations need to look for efficient data architecture solutions.

Reducing cost while accelerating Cassandra performance for a product catalog use case

As Cassandra is built to scale with the flexible insertion of multiple nodes to any cluster, adding nodes is an easy solution.  Cassandra is the database of choice for many retailers and eCommerce organizations, and in this use case, is the transactional system for a product catalog.

product catalog_before

This eCommerce Giant had already tuned existing nodes to optimize for performance and employed a highly efficient compression strategy, but still found themselves adding more database nodes and expanding cloud infrastructure. 

The issue with continuous tuning, adding of servers or cloud instances, or expanding nodes is that cost also rises with each, which was the case for this eCommerce Giant, who saw a continuously rising (and unmanageable) Google Cloud Platform bill.

The primary goals for this eCommerce giant while looking for solutions were:

  • Reduce latency to meet SLA windows, especially during peak traffic times
  • Increase performance as much as possible for existing clusters
  • Cut TCO or plateau cloud costs

This combination of needs is precisely what drove this eCommerce giant to look for a solution that would boost existing Apache Cassandra and DataStax node performance while cutting costs by at least 30%.

High level reference architecture: Cassandra accelerated by rENIAC for a product catalog

Product catalog_after


rENIAC can be deployed as a managed cache in the cloud to provide users with immediate performance improvements. Because rENIAC is CQL compatible, it served as a drop-in addition to the eCommerce Giant’s existing Apache Cassandra (DataStax Enterprise) database clusters. 

The choice of this eCommerce company to test rENIAC as a managed cache means users are not responsible for administering it as a separate technology, but merely spinning up a cloud instance with rENIAC and allowing it to intelligently store and route queries with Apache Cassandra.  As a managed cache, Cassandra updates the data stored in the rENIAC cache and rENIAC provides a little over 1TB of data (copied two additional times for replication purposes) on a single instance.

48% reduction in cloud instance cost 

The benefits this eCommerce Giant realized while deploying rENIAC as a cache are:

  • 2.9x Improvement in throughput
  • 2.6x improvement in mean latency 
  • 10.8x improvement in p99 latency, i.e. elimination of long tail latency 
  • 48% reduction in cloud instance cost - 18% more than the savings goal

Download the full case study including benchmark results below.

Download the full case study